SurveyUSA admits they screwed up VA Gov poll
By Chad Dotson Posted in Elections — Comments (9) / Email this page » / Leave a comment »
In the comments below and on some lefty blogs, there has been much discussion of a SurveyUSA poll today showing Democrat Tim Kaine with a 9 point lead over Republican Jerry Kilgore in the Virginia Governor's race.
I looked at the crosstabs, and immediately noticed about a dozen screwy numbers, starting with the fact that Democrats were way oversampled.
Well, SurveyUSA has now admitted that their poll was screwed up. For all intents and purposes, the race for Governor remains a dead heat:
Read on....
UNUSUAL VOLATILITY IN VA AS VOTERS GO TO BED ON ELECTION EVE: Interviews in the Virginia governor's race conducted by SurveyUSA tonight Monday 11/7 (but before President Bush appeared in Richmond) show a swing back towards Republican candidate Jerry Kilgore, causing SurveyUSA to now update its final projection in the Virginia Governor's Contest. This morning, based on interviews conducted Friday, Saturday and Sunday (11/4/05 through 11/6/05), SurveyUSA released data that showed Democrat Tim Kaine 9 points ahead of Kilgore. However, because of intra-day volatility in that data, SurveyUSA continued to poll throughout the afternoon and evening today Monday 11/7. When interviews from the most recent 3 days -- Saturday, Sunday and today Monday -- are averaged, Kaine's lead shrinks now to 5 points. When interviews from just the past two days -- Sunday and today Monday -- are averaged, the contest is closer yet. When interviews from Monday only are considered, the contest is tied, but the Margin of Sampling error from just the one day of interviewing is high enough, and the results aberrant enough, that SurveyUSA is uncomfortable reporting just Monday-only data. For the record, SurveyUSA goes into the clubhouse with its final projection (based on Saturday, Sunday and Monday polling): Kaine 50%, Kilgore 45%. A closer outcome still is possible.
We're still within the margin of error. I stand by my predictions from earlier. Heck, even with these results, you have to remember that SurveyUSA has a reputation for polls that lean left.
This is just another reason why Virginia Republicans need to ignore these polls at this late date and work on getting people to the polls. This race is tied, which means it's very winnable for Jerry Kilgore.
Heck, based on what I've seen of the Republican GOTV machine -- and what I'm beginning to hear about the outstanding Kilgore rally with President Bush tonight -- there are many reasons to be very optimistic about tomorrow's results.
« Question and answer time: the Wes Clark thing. — Comments (50) | Election Day VA/NJ Open Thread — Comments (55) »
SurveyUSA admits they screwed up VA Gov poll 9 Comments (0 topical, 9 editorial, 0 hidden) Post a comment »
because they can easily be purposefully manipulated and the average person isn't going care about or understand the importance of the internals, sample and methodology. They are just going to see the numbers.
Did the rally go well and was it covered in the VA media? Anyone from VA able to respond?
come up with any honest reason for how SurveyUSA could give any Democrat that kind of a lead in Virginia.
What's the quote from the NYTimes in 1972? 'I don't know how Nixon could have won, everyone I know voted for McGovern.'
Can anyone recall a poll put out by an MSM outlet (e.g. newspaper, television news) right before an election which "oversampled" likely Republican voters?
It's unlikely that there would be a legitimate reason to oversample on any party dimension.
A survey will often oversample in order to sub-group analysis. Let's say you're interested in the opinions of Jewish voters. A standard sample of 800 people on a national survey might yield 30 or 40 Jewish voters, which isn't enough to do any reliable analysis of that group. So, you oversample to bring that number up. When you calculate your overall results, you factor in the oversample so that it doesn't distort the results.
The reason why you wouldn't oversample on Republicans or Democrats is that a sample of 800 will yield at least 200 of each party, which is sufficient for some sub-group analysis.
Well, gee. Only off by 1%. Seems pretty close to me!
But it's more like 52-46 - still pretty close to the prediction.
BTW folks, with 1% of the precincts yet to report the Democrat has taken a roughly 1,400-vote (0.1%) lead in the AG's race - which will almost certainly be recounted.
Interestingly, the GOP AG nominee outpolled Kilgore (top of the ticket) by nearly 60k-votes. If he loses, it could be argued very credibly that it's because of Kilgore's anemic performance.
editorials, what's the differnce? The MSM knows that todays modern video renter, they used to be called citizens, will be influenced by poll results. As incomprehensible as it is people are swayed by by what unknown other video renters think,a sort of mystical, remote control peer pressure that temporarily tickles the brain as they pop one video out and slide another one in. It is not unusual for democrats to be over represented in polls but we must not get upset and harbor dark thoughts about dishonesty, it's all just good, clean fun and Mama Media knows best.